Instructions to use OptimizeLLM/Qwen3.5-122B-A10B-heretic-MTP-NVFP4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use OptimizeLLM/Qwen3.5-122B-A10B-heretic-MTP-NVFP4 with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="OptimizeLLM/Qwen3.5-122B-A10B-heretic-MTP-NVFP4")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("OptimizeLLM/Qwen3.5-122B-A10B-heretic-MTP-NVFP4")
model = AutoModelForMultimodalLM.from_pretrained("OptimizeLLM/Qwen3.5-122B-A10B-heretic-MTP-NVFP4", device_map="auto")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
inputs = processor.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use OptimizeLLM/Qwen3.5-122B-A10B-heretic-MTP-NVFP4 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "OptimizeLLM/Qwen3.5-122B-A10B-heretic-MTP-NVFP4"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "OptimizeLLM/Qwen3.5-122B-A10B-heretic-MTP-NVFP4",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/OptimizeLLM/Qwen3.5-122B-A10B-heretic-MTP-NVFP4

SGLang

How to use OptimizeLLM/Qwen3.5-122B-A10B-heretic-MTP-NVFP4 with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "OptimizeLLM/Qwen3.5-122B-A10B-heretic-MTP-NVFP4" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "OptimizeLLM/Qwen3.5-122B-A10B-heretic-MTP-NVFP4",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "OptimizeLLM/Qwen3.5-122B-A10B-heretic-MTP-NVFP4" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "OptimizeLLM/Qwen3.5-122B-A10B-heretic-MTP-NVFP4",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use OptimizeLLM/Qwen3.5-122B-A10B-heretic-MTP-NVFP4 with Docker Model Runner:
```
docker model run hf.co/OptimizeLLM/Qwen3.5-122B-A10B-heretic-MTP-NVFP4
```

Qwen3.5-122B-A10B-heretic-MTP-NVFP4

NVFP4 (W4A4) quantization of trohrbaugh/Qwen3.5-122B-A10B-heretic.

Base Model: Qwen3.5-122B-A10B-heretic (MoE: 122B total, ~10B active, abliterated with KL ~0.09)
Quantization: NVFP4 (W4A4) — weights and activations
Size: 76GB (16 shards + MTP shard + visual shard)
Quantized: Language backbone MoE expert and attention layers
NOT Quantized: Vision encoder, merger, LM head, embed tokens, linear attention, MoE gates, MTP heads (remain BF16)
MTP: Working speculative decoding — 785 tensors spliced from base Qwen/Qwen3.5-122B-A10B in BF16
Tokenizer: From base Qwen3.5

Usage with vLLM

Tested on vLLM 0.19+.

vllm serve OptimizeLLM/Qwen3.5-122B-A10B-heretic-MTP-NVFP4 \
  --tensor-parallel-size 2 \
  --reasoning-parser qwen3 \
  --reasoning-config '{}' \
  --enable-auto-tool-choice \
  --tool-call-parser qwen3_coder \
  --enable-prefix-caching \
  --speculative-config '{"method":"mtp","num_speculative_tokens":6}'

MTP Throughput (2x RTX 6000 Pro Blackwell, TP=2)

MTP Speculative Tokens	tok/s
0 (disabled)	~105
1	~115
2	~145
3	~170
6	~190

Quantization

Quantized with llm-compressor (compressed-tensors v0.14.1.dev28).

Format: nvfp4-pack-quantized
Weight/Activation bits: FP4 E2M1
Scale dtype: float8_e4m3fn
Group size: 16
Calibration: 512 samples (256 UltraChat + 256 Nemotron-CC chat split)
MoE calibration: moe_calibrate_all_experts=True
Ignore list: Aligned with RedHatAI/Qwen3.5-122B-A10B-NVFP4

MTP Splicing

The upstream heretic quant quantized MTP heads to FP4, which breaks speculative decoding. All 785 mtp.* tensors were extracted from base Qwen/Qwen3.5-122B-A10B in BF16 and saved as a separate model_mtp.safetensors shard (4.8GB).

Notes for Reproducing

MoE unfusing roughly doubles memory per layer during calibration. GPU calibration will likely OOM. CPU-only calibration with ~400GB swap worked (peak ~445GB RSS+swap, ~3 days on 64 cores).
llm-compressor's use_auth_token kwarg crashes on transformers 5.x — patch to token.
offload_folder is required even without disk offload (transformers reformats fused expert tensors during loading).
ctypes.CDLL("libc.so.6").mallopt(-1, 0) before loading prevents glibc malloc arena bloat.

Acknowledgments

Qwen Team for Qwen3.5-122B-A10B
trohrbaugh for the heretic abliteration
RedHatAI for the reference NVFP4 recipe
vLLM and llm-compressor teams

Downloads last month: 5,918

Safetensors

Model size

74B params

Tensor type

F32

BF16

F8_E4M3

Model tree for OptimizeLLM/Qwen3.5-122B-A10B-heretic-MTP-NVFP4

Base model

Qwen/Qwen3.5-122B-A10B

Quantized

(151)

this model